Claude Mods: Flipping the Harness Inside Out
Short summary
Claude Mods are TypeScript or JavaScript plugins that hook into Claude Code's events, letting you control when the agent interrupts you and what it shows you. The author compares this approach with ChatGPT's Intelligent UI and Cursor's PR Routing, arguing that Mods let the user design the interface rather than accept one a company built.
- Mods are middleware that intercept prompts, tool calls, and renders, so you can change how Claude Code behaves and displays information at runtime.
- The built-in You Should Know mod caught uncommitted work in 45 files that a demo edit would have discarded, showing how a side agent can flag what the main agent buries.
- The author's Next Steps Supervisor fired on 99 of 111 turns, so the noise and token cost (roughly 5–6k tokens per check) undercut its value, and mods are unsandboxed code running with your permissions.
Full summary
Claude Mods are plugins for Claude Code that run TypeScript or JavaScript functions inside the CLI and Desktop app. They hook into events like prompts, tool calls, and interface renders, so they can observe those events, change how they display, or replace the default handling entirely. The author's framing is that the agent harness is shifting from a pipe between the model and the machine into a layer that faces the human, which he calls the "Attention Interface": the part that decides when the agent interrupts you, what it shows you, and which decisions it hands back. Mods make that layer programmable by the person using it, and Claude can write them for you from plain-language requests thanks to a built-in plugin-authoring skill.
Anthropic already ships six built-in mods, visible through /plugin under "Built-in." They include agents-md, which loads project instructions; diff, which takes over the /diff pane; sec-default, which guards organization-managed settings from user-installed mods; telemetry; plugin-authoring; and you-should-know, the only one off by default. The author's favorite is You Should Know, a side agent that watches the main agent and surfaces things you might miss. In one case it caught that CompareView.swift and about 45 other files already had uncommitted work, which would have been lost if the agent's demo edits were discarded. He also built Next Steps Supervisor, which runs after turns that edit files, run commands, or make eight or more tool calls, and offers clickable follow-up actions. It ships in his open-source ModSmith plugin.
The piece then compares Claude Mods with two other attempts at the same problem. OpenAI's ChatGPT Intelligent UI, launched October 7 alongside GPT-6, lets the model generate charts, forms, and small tools per answer from a component library OpenAI controls, with only a visuals toggle offered as user control. Cursor's PR Routing & Approval works at the team level, assigning reviewers by code ownership and history, auto-approving low-risk PRs against a configurable threshold, and blocking agents from relaxing policy by reading approval rules from the base branch. Cursor's own docs say it "does not replace a full code review." The author's distinction is about who designs the interface: the model in ChatGPT, a fixed policy workflow in Cursor, and the user in Claude Mods, which is the only one of the three where the person shapes the attention layer directly.